CSAIL Publications and Digital Archive header
bullet Research Abstracts Home bullet CSAIL Digital Archive bullet Research Activities bullet CSAIL Home bullet

link to publications.csail.mit.edu link to www.csail.mit.edu horizontal line


Research Abstracts - 2007
horizontal line

horizontal line

vertical line
vertical line

Control of Simulated Creatures

Yeuhi Abe, Marco da Silva & Jovan Popović


Animated characters, such as humans, animals, and robots, are typically at the center of attention in the virtual worlds they inhabit. The quality of their motion is one of the most important details in compelling animation, especially for interactive applications such as computer games and training simulations. Characters that move in response to their environment with intention and grace will enhance applications by immersing viewers in a palpable experience. In contrast, characters that move in an artificial or scripted manner will detract from the realism of the virtual world, making applications less effective.

A general solution to "on the fly" synthesis of character animations for interactive applications has yet to be discovered. This is because interactivity significantly complicates the task of generating character animations. Unlike static animation (e.g., in movies), interactive characters must react in realistic and intelligent ways in response to the actions of the user and the surrounding environment.

Physical simulation offers one promising approach to resolving this problem. But the use of physical simulation for the animation of active characters has stalled due to the significant difficulty of developing control policies that compute the equivalent of muscle forces in Nature, which guide the characters through desired motion sequences. As a result, most current animation systems either simulate characters as passive rag dolls or play back pre-recorded motions that ignore dynamics completely. Neither solution allows for the level of realism that will be expected from the next generation of interactive applications.


We have developed two new control algorithm: prioritized control [1] and multiobjective control [2]. Both simplify the process of directing physically simulated characters by decoupling the description of motion from the computation of physical control parameters. Prioritized control is faster, but cannot handle unilateral contact with the environment. Multiobjective control is slightly slower but it allows for handling of arbitrary unilateral contact with the environment as well as more flexibility in motion specification.

Object Manipulation

Using prioritized control we have created animations of characters lifting, tossing, and catching objects. Since the characters are physically simulated, dynamic motion effects emerge naturally, such as a proper response to the weight of the object being manipulated and reactions to unexpected collisions with objects. For example, an animation of a character lifting a heavy object will proceed slower than the same animation for a lighter object due to explicit limits on the force the character is allowed to use to lift the object. (See Figure 1) Similar effects are observed when manipulating fast moving objects or when force disturbances are added while the character is holding an object.

Figure 1: Different animations of a character lifting a box (top row) emerge automatically when the weight of the box is increase (middle row) or the posture model is changed (bottom row).
Character lifting a box.
Dynamic Balancing

Using multiobjective control we have created several interactive simulations of a character executing motions while maintaining balance, despite significant disturbances in the environment. (See Figure 2) These disturbances include collisions with moving objects, forces applied by a user, and random motions of a moving platform on which the character is standing.

Characters balancing.

Figure 2: These images are snapshots taken from interactive simulations driven by our control algorithm. The articulated human body tracks motion capture data while maintaining balance.

Future Work

Our goal is to control the animation of lifelike characters as they accomplish even more complex dynamic manipulation tasks such as dribbling a basketball, interacting with each other, or locomoting. Previous work has shown how to control complex simulation of passive objects (e.g., simple rigid bodies like scissors. See figure 3.) so that they meet animator specified constraints [3]. We hope to build upon this work to control both passive objects and active characters simultaneously within our simulation and control framework.

Figure 3: A bounce and a flip is added to an animation where scissors lands on a coat tack.
Animation of scissors.

[1] Yeuhi Abe and Jovan Popović. Interactive Animation of Dynamic Manipulation. ACM/EUROGRAPHICS Symposium on Computer Animation. 2004.

[2] Yeuhi Abe. (2007) Towards Adaptive and Directable Control of Simulated Creatures. Masters Thesis, MIT, February 2007.

[3] Jovan Popović, Steven M. Seitz, Michael Erdmann, Zoran Popović, Andrew Witkin. Interactive Manipulation of Rigid Body Simulations. In Computer Graphics (Proceedings of SIGGRAPH 2000), ACM SIGGRAPH, Annual Conference Series, pp. 209-217.


vertical line
vertical line
horizontal line

MIT logo Computer Science and Artificial Intelligence Laboratory (CSAIL)
The Stata Center, Building 32 - 32 Vassar Street - Cambridge, MA 02139 - USA
tel:+1-617-253-0073 - publications@csail.mit.edu